SkyReels Hunyuan T2V & I2V (#10837)

* update

* make fix-copies

* update

* tests

* update

* update

* add co-author
Co-Authored-By: Langdx <82783347+Langdx@users.noreply.github.com>

* add co-author
Co-Authored-By: howe <howezhang2018@gmail.com>

* update

---------
Co-authored-by: Langdx <82783347+Langdx@users.noreply.github.com>
Co-authored-by: howe <howezhang2018@gmail.com>

docs/source/en/api/pipelines/hunyuan_video.md

@@ -32,6 +32,21 @@ Recommendations for inference:
 - For smaller resolution videos, try lower values of `shift` (between `2.0` to `5.0`) in the [Scheduler](https://huggingface.co/docs/diffusers/main/en/api/schedulers/flow_match_euler_discrete#diffusers.FlowMatchEulerDiscreteScheduler.shift). For larger resolution images, try higher values (between `7.0` and `12.0`). The default value is `7.0` for HunyuanVideo.
 - For more information about supported resolutions and other details, please refer to the original repository [here](https://github.com/Tencent/HunyuanVideo/).
 
+## Available models
+
+The following models are available for the [`HunyuanVideoPipeline`](text-to-video) pipeline:
+
+| Model name | Description |
+|:---|:---|
+| [`hunyuanvideo-community/HunyuanVideo`](https://huggingface.co/hunyuanvideo-community/HunyuanVideo) | Official HunyuanVideo (guidance-distilled). Performs best at multiple resolutions and frames. Performs best with `guidance_scale=6.0`, `true_cfg_scale=1.0` and without a negative prompt. |
+| [`https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-T2V`](https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-T2V) | Skywork's custom finetune of HunyuanVideo (de-distilled). Performs best with `97x544x960` resolution, `guidance_scale=1.0`, `true_cfg_scale=6.0` and a negative prompt. |
+
+The following models are available for the image-to-video pipeline:
+
+| Model name | Description |
+|:---|:---|
+| [`https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-I2V`](https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-I2V) | Skywork's custom finetune of HunyuanVideo (de-distilled). Performs best with `97x544x960` resolution. Performs best at `97x544x960` resolution, `guidance_scale=1.0`, `true_cfg_scale=6.0` and a negative prompt. |
+
 ## Quantization
 
 Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.

src/diffusers/__init__.py

@@ -305,6 +305,7 @@ else:
             "HunyuanDiTControlNetPipeline",
             "HunyuanDiTPAGPipeline",
             "HunyuanDiTPipeline",
+            "HunyuanSkyreelsImageToVideoPipeline",
             "HunyuanVideoPipeline",
             "I2VGenXLPipeline",
             "IFImg2ImgPipeline",
@@ -804,6 +805,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             HunyuanDiTControlNetPipeline,
             HunyuanDiTPAGPipeline,
             HunyuanDiTPipeline,
+            HunyuanSkyreelsImageToVideoPipeline,
             HunyuanVideoPipeline,
             I2VGenXLPipeline,
             IFImg2ImgPipeline,

src/diffusers/pipelines/__init__.py

@@ -217,7 +217,7 @@ else:
         "IFSuperResolutionPipeline",
     ]
     _import_structure["hunyuandit"] = ["HunyuanDiTPipeline"]
-    _import_structure["hunyuan_video"] = ["HunyuanVideoPipeline"]
+    _import_structure["hunyuan_video"] = ["HunyuanVideoPipeline", "HunyuanSkyreelsImageToVideoPipeline"]
     _import_structure["kandinsky"] = [
         "KandinskyCombinedPipeline",
         "KandinskyImg2ImgCombinedPipeline",
@@ -558,7 +558,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             FluxPriorReduxPipeline,
             ReduxImageEncoder,
         )
-        from .hunyuan_video import HunyuanVideoPipeline
+        from .hunyuan_video import HunyuanSkyreelsImageToVideoPipeline, HunyuanVideoPipeline
         from .hunyuandit import HunyuanDiTPipeline
         from .i2vgen_xl import I2VGenXLPipeline
         from .kandinsky import (

src/diffusers/pipelines/hunyuan_video/__init__.py

@@ -22,6 +22,7 @@ except OptionalDependencyNotAvailable:
 
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
+    _import_structure["pipeline_hunyuan_skyreels_image2video"] = ["HunyuanSkyreelsImageToVideoPipeline"]
     _import_structure["pipeline_hunyuan_video"] = ["HunyuanVideoPipeline"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
@@ -32,6 +33,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     except OptionalDependencyNotAvailable:
         from ...utils.dummy_torch_and_transformers_objects import *
     else:
+        from .pipeline_hunyuan_skyreels_image2video import HunyuanSkyreelsImageToVideoPipeline
         from .pipeline_hunyuan_video import HunyuanVideoPipeline
 
 else:

src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_skyreels_image2video.py

+# Copyright 2024 The HunyuanVideo Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer, LlamaModel, LlamaTokenizerFast
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput
+from ...loaders import HunyuanVideoLoraLoaderMixin
+from ...models import AutoencoderKLHunyuanVideo, HunyuanVideoTransformer3DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import HunyuanVideoPipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```python
+        >>> import torch
+        >>> from diffusers import HunyuanSkyreelsImageToVideoPipeline, HunyuanVideoTransformer3DModel
+        >>> from diffusers.utils import load_image, export_to_video
+
+        >>> model_id = "hunyuanvideo-community/HunyuanVideo"
+        >>> transformer_model_id = "Skywork/SkyReels-V1-Hunyuan-I2V"
+        >>> transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+        ...     transformer_model_id, torch_dtype=torch.bfloat16
+        ... )
+        >>> pipe = HunyuanSkyreelsImageToVideoPipeline.from_pretrained(
+        ...     model_id, transformer=transformer, torch_dtype=torch.float16
+        ... )
+        >>> pipe.vae.enable_tiling()
+        >>> pipe.to("cuda")
+
+        >>> prompt = "An astronaut hatching from an egg, on the surface of the moon, the darkness and depth of space realised in the background. High quality, ultrarealistic detail and breath-taking movie-like camera shot."
+        >>> negative_prompt = "Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion"
+        >>> image = load_image(
+        ...     "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
+        ... )
+
+        >>> output = pipe(
+        ...     image=image,
+        ...     prompt=prompt,
+        ...     negative_prompt=negative_prompt,
+        ...     num_inference_steps=30,
+        ...     true_cfg_scale=6.0,
+        ...     guidance_scale=1.0,
+        ... ).frames[0]
+        >>> export_to_video(output, "output.mp4", fps=15)
+        ```
+"""
+
+
+DEFAULT_PROMPT_TEMPLATE = {
+    "template": (
+        "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
+        "1. The main content and theme of the video."
+        "2. The color, shape, size, texture, quantity, text, and spatial relationships of the objects."
+        "3. Actions, events, behaviors temporal relationships, physical movement changes of the objects."
+        "4. background environment, light, style and atmosphere."
+        "5. camera angles, movements, and transitions used in the video:<|eot_id|>"
+        "<|start_header_id|>user<|end_header_id|>\n\n{}<|eot_id|>"
+    ),
+    "crop_start": 95,
+}
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+    encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+    if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+        return encoder_output.latent_dist.sample(generator)
+    elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+        return encoder_output.latent_dist.mode()
+    elif hasattr(encoder_output, "latents"):
+        return encoder_output.latents
+    else:
+        raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class HunyuanSkyreelsImageToVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
+    r"""
+    Pipeline for image-to-video generation using HunyuanVideo.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    Args:
+        text_encoder ([`LlamaModel`]):
+            [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        tokenizer (`LlamaTokenizer`):
+            Tokenizer from [Llava Llama3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).
+        transformer ([`HunyuanVideoTransformer3DModel`]):
+            Conditional Transformer to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKLHunyuanVideo`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+        text_encoder_2 ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer_2 (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        text_encoder: LlamaModel,
+        tokenizer: LlamaTokenizerFast,
+        transformer: HunyuanVideoTransformer3DModel,
+        vae: AutoencoderKLHunyuanVideo,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        text_encoder_2: CLIPTextModel,
+        tokenizer_2: CLIPTokenizer,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+        )
+
+        self.vae_scale_factor_temporal = self.vae.temporal_compression_ratio if getattr(self, "vae", None) else 4
+        self.vae_scale_factor_spatial = self.vae.spatial_compression_ratio if getattr(self, "vae", None) else 8
+        self.vae_scaling_factor = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.476986
+        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_llama_prompt_embeds
+    def _get_llama_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_template: Dict[str, Any],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+        num_hidden_layers_to_skip: int = 2,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        prompt = [prompt_template["template"].format(p) for p in prompt]
+
+        crop_start = prompt_template.get("crop_start", None)
+        if crop_start is None:
+            prompt_template_input = self.tokenizer(
+                prompt_template["template"],
+                padding="max_length",
+                return_tensors="pt",
+                return_length=False,
+                return_overflowing_tokens=False,
+                return_attention_mask=False,
+            )
+            crop_start = prompt_template_input["input_ids"].shape[-1]
+            # Remove <|eot_id|> token and placeholder {}
+            crop_start -= 2
+
+        max_sequence_length += crop_start
+        text_inputs = self.tokenizer(
+            prompt,
+            max_length=max_sequence_length,
+            padding="max_length",
+            truncation=True,
+            return_tensors="pt",
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_attention_mask=True,
+        )
+        text_input_ids = text_inputs.input_ids.to(device=device)
+        prompt_attention_mask = text_inputs.attention_mask.to(device=device)
+
+        prompt_embeds = self.text_encoder(
+            input_ids=text_input_ids,
+            attention_mask=prompt_attention_mask,
+            output_hidden_states=True,
+        ).hidden_states[-(num_hidden_layers_to_skip + 1)]
+        prompt_embeds = prompt_embeds.to(dtype=dtype)
+
+        if crop_start is not None and crop_start > 0:
+            prompt_embeds = prompt_embeds[:, crop_start:]
+            prompt_attention_mask = prompt_attention_mask[:, crop_start:]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        _, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+        prompt_attention_mask = prompt_attention_mask.repeat(1, num_videos_per_prompt)
+        prompt_attention_mask = prompt_attention_mask.view(batch_size * num_videos_per_prompt, seq_len)
+
+        return prompt_embeds, prompt_attention_mask
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline._get_clip_prompt_embeds
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_videos_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 77,
+    ) -> torch.Tensor:
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder_2.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False).pooler_output
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt)
+        prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, -1)
+
+        return prompt_embeds
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]] = None,
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        num_videos_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+        max_sequence_length: int = 256,
+    ):
+        if prompt_embeds is None:
+            prompt_embeds, prompt_attention_mask = self._get_llama_prompt_embeds(
+                prompt,
+                prompt_template,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=max_sequence_length,
+            )
+
+        if pooled_prompt_embeds is None:
+            if prompt_2 is None:
+                prompt_2 = prompt
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt,
+                num_videos_per_prompt,
+                device=device,
+                dtype=dtype,
+                max_sequence_length=77,
+            )
+
+        return prompt_embeds, pooled_prompt_embeds, prompt_attention_mask
+
+    # Copied from diffusers.pipelines.hunyuan_video.pipeline_hunyuan_video.HunyuanVideoPipeline.check_inputs
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        prompt_template=None,
+    ):
+        if height % 16 != 0 or width % 16 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 16 but are {height} and {width}.")
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_template is not None:
+            if not isinstance(prompt_template, dict):
+                raise ValueError(f"`prompt_template` has to be of type `dict` but is {type(prompt_template)}")
+            if "template" not in prompt_template:
+                raise ValueError(
+                    f"`prompt_template` has to contain a key `template` but only found {prompt_template.keys()}"
+                )
+
+    def prepare_latents(
+        self,
+        image: torch.Tensor,
+        batch_size: int,
+        num_channels_latents: int = 32,
+        height: int = 544,
+        width: int = 960,
+        num_frames: int = 97,
+        dtype: Optional[torch.dtype] = None,
+        device: Optional[torch.device] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        image = image.unsqueeze(2)  # [B, C, 1, H, W]
+        if isinstance(generator, list):
+            image_latents = [
+                retrieve_latents(self.vae.encode(image[i].unsqueeze(0)), generator[i]) for i in range(batch_size)
+            ]
+        else:
+            image_latents = [retrieve_latents(self.vae.encode(img.unsqueeze(0)), generator) for img in image]
+
+        image_latents = torch.cat(image_latents, dim=0).to(dtype) * self.vae_scaling_factor
+
+        num_latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+        latent_height, latent_width = height // self.vae_scale_factor_spatial, width // self.vae_scale_factor_spatial
+        shape = (batch_size, num_channels_latents, num_latent_frames, latent_height, latent_width)
+        padding_shape = (batch_size, num_channels_latents, num_latent_frames - 1, latent_height, latent_width)
+
+        latents_padding = torch.zeros(padding_shape, dtype=dtype, device=device)
+        image_latents = torch.cat([image_latents, latents_padding], dim=2)
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, dtype=dtype, device=device)
+        else:
+            latents = latents.to(dtype=dtype, device=device)
+
+        return latents, image_latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        image: PipelineImageInput,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
+        height: int = 544,
+        width: int = 960,
+        num_frames: int = 97,
+        num_inference_steps: int = 50,
+        sigmas: List[float] = None,
+        true_cfg_scale: float = 6.0,
+        guidance_scale: float = 1.0,
+        num_videos_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[
+            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+        ] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        prompt_template: Dict[str, Any] = DEFAULT_PROMPT_TEMPLATE,
+        max_sequence_length: int = 256,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
+            height (`int`, defaults to `720`):
+                The height in pixels of the generated image.
+            width (`int`, defaults to `1280`):
+                The width in pixels of the generated image.
+            num_frames (`int`, defaults to `129`):
+                The number of frames in the generated video.
+            num_inference_steps (`int`, defaults to `50`):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+            guidance_scale (`float`, defaults to `6.0`):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality. Note that the only available HunyuanVideo model is
+                CFG-distilled, which means that traditional guidance between unconditional and conditional latent is
+                not applied.
+            num_videos_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.Tensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`HunyuanVideoPipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~HunyuanVideoPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`HunyuanVideoPipelineOutput`] is returned, otherwise a `tuple` is returned
+                where the first element is a list with the generated images and the second element is a list of `bool`s
+                indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content.
+        """
+
+        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds,
+            callback_on_step_end_tensor_inputs,
+            prompt_template,
+        )
+
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        device = self._execution_device
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        # 3. Encode input prompt
+        transformer_dtype = self.transformer.dtype
+        prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_template=prompt_template,
+            num_videos_per_prompt=num_videos_per_prompt,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            device=device,
+            max_sequence_length=max_sequence_length,
+        )
+        prompt_embeds = prompt_embeds.to(transformer_dtype)
+        prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
+
+        # 4. Prepare timesteps
+        sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
+
+        # 5. Prepare latent variables
+        vae_dtype = self.vae.dtype
+        image = self.video_processor.preprocess(image, height=height, width=width).to(device, vae_dtype)
+        num_channels_latents = self.transformer.config.in_channels // 2
+        latents, image_latents = self.prepare_latents(
+            image,
+            batch_size * num_videos_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            num_frames,
+            torch.float32,
+            device,
+            generator,
+            latents,
+        )
+        latent_image_input = image_latents.to(transformer_dtype)
+
+        # 6. Prepare guidance condition
+        guidance = torch.tensor([guidance_scale] * latents.shape[0], dtype=transformer_dtype, device=device) * 1000.0
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                self._current_timestep = t
+                latent_model_input = latents.to(transformer_dtype)
+                latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=1)
+
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                noise_pred = self.transformer(
+                    hidden_states=latent_model_input,
+                    timestep=timestep,
+                    encoder_hidden_states=prompt_embeds,
+                    encoder_attention_mask=prompt_attention_mask,
+                    pooled_projections=pooled_prompt_embeds,
+                    guidance=guidance,
+                    attention_kwargs=attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                if do_true_cfg:
+                    neg_noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        timestep=timestep,
+                        encoder_hidden_states=negative_prompt_embeds,
+                        encoder_attention_mask=negative_prompt_attention_mask,
+                        pooled_projections=negative_pooled_prompt_embeds,
+                        guidance=guidance,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+                    noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        self._current_timestep = None
+
+        if not output_type == "latent":
+            latents = latents.to(self.vae.dtype) / self.vae.config.scaling_factor
+            video = self.vae.decode(latents, return_dict=False)[0]
+            video = self.video_processor.postprocess_video(video, output_type=output_type)
+        else:
+            video = latents
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return HunyuanVideoPipelineOutput(frames=video)

src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py

@@ -325,7 +325,7 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             )
 
         if pooled_prompt_embeds is None:
-            if prompt_2 is None and pooled_prompt_embeds is None:
+            if prompt_2 is None:
                 prompt_2 = prompt
             pooled_prompt_embeds = self._get_clip_prompt_embeds(
                 prompt,
@@ -470,11 +470,14 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
         self,
         prompt: Union[str, List[str]] = None,
         prompt_2: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        negative_prompt_2: Union[str, List[str]] = None,
         height: int = 720,
         width: int = 1280,
         num_frames: int = 129,
         num_inference_steps: int = 50,
         sigmas: List[float] = None,
+        true_cfg_scale: float = 1.0,
         guidance_scale: float = 6.0,
         num_videos_per_prompt: Optional[int] = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
@@ -482,6 +485,9 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
         prompt_embeds: Optional[torch.Tensor] = None,
         pooled_prompt_embeds: Optional[torch.Tensor] = None,
         prompt_attention_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_attention_mask: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -502,6 +508,13 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             prompt_2 (`str` or `List[str]`, *optional*):
                 The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                 will be used instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
             height (`int`, defaults to `720`):
                 The height in pixels of the generated image.
             width (`int`, defaults to `1280`):
@@ -515,6 +528,8 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
                 Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
             guidance_scale (`float`, defaults to `6.0`):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
@@ -535,6 +550,17 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             prompt_embeds (`torch.Tensor`, *optional*):
                 Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                 provided, text embeddings are generated from the `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
             output_type (`str`, *optional*, defaults to `"pil"`):
                 The output format of the generated image. Choose between `PIL.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
@@ -579,6 +605,11 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             prompt_template,
         )
 
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_pooled_prompt_embeds is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+
         self._guidance_scale = guidance_scale
         self._attention_kwargs = attention_kwargs
         self._current_timestep = None
@@ -595,6 +626,7 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             batch_size = prompt_embeds.shape[0]
 
         # 3. Encode input prompt
+        transformer_dtype = self.transformer.dtype
         prompt_embeds, pooled_prompt_embeds, prompt_attention_mask = self.encode_prompt(
             prompt=prompt,
             prompt_2=prompt_2,
@@ -606,21 +638,29 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
             device=device,
             max_sequence_length=max_sequence_length,
         )
-
-        transformer_dtype = self.transformer.dtype
         prompt_embeds = prompt_embeds.to(transformer_dtype)
         prompt_attention_mask = prompt_attention_mask.to(transformer_dtype)
-        if pooled_prompt_embeds is not None:
-            pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+        pooled_prompt_embeds = pooled_prompt_embeds.to(transformer_dtype)
+
+        if do_true_cfg:
+            negative_prompt_embeds, negative_pooled_prompt_embeds, negative_prompt_attention_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_2=negative_prompt_2,
+                prompt_template=prompt_template,
+                num_videos_per_prompt=num_videos_per_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                prompt_attention_mask=negative_prompt_attention_mask,
+                device=device,
+                max_sequence_length=max_sequence_length,
+            )
+            negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(transformer_dtype)
+            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(transformer_dtype)
 
         # 4. Prepare timesteps
         sigmas = np.linspace(1.0, 0.0, num_inference_steps + 1)[:-1] if sigmas is None else sigmas
-        timesteps, num_inference_steps = retrieve_timesteps(
-            self.scheduler,
-            num_inference_steps,
-            device,
-            sigmas=sigmas,
-        )
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
 
         # 5. Prepare latent variables
         num_channels_latents = self.transformer.config.in_channels
@@ -664,6 +704,19 @@ class HunyuanVideoPipeline(DiffusionPipeline, HunyuanVideoLoraLoaderMixin):
                     return_dict=False,
                 )[0]
 
+                if do_true_cfg:
+                    neg_noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        timestep=timestep,
+                        encoder_hidden_states=negative_prompt_embeds,
+                        encoder_attention_mask=negative_prompt_attention_mask,
+                        pooled_projections=negative_pooled_prompt_embeds,
+                        guidance=guidance,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
+                    noise_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
                 # compute the previous noisy sample x_t -> x_t-1
                 latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -617,6 +617,21 @@ class HunyuanDiTPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class HunyuanSkyreelsImageToVideoPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class HunyuanVideoPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

tests/models/transformers/test_models_transformer_hunyuan_video.py

@@ -87,3 +87,70 @@ class HunyuanVideoTransformer3DTests(ModelTesterMixin, unittest.TestCase):
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"HunyuanVideoTransformer3DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
+
+
+class HunyuanSkyreelsImageToVideoTransformer3DTests(ModelTesterMixin, unittest.TestCase):
+    model_class = HunyuanVideoTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
+
+    @property
+    def dummy_input(self):
+        batch_size = 1
+        num_channels = 8
+        num_frames = 1
+        height = 16
+        width = 16
+        text_encoder_embedding_dim = 16
+        pooled_projection_dim = 8
+        sequence_length = 12
+
+        hidden_states = torch.randn((batch_size, num_channels, num_frames, height, width)).to(torch_device)
+        timestep = torch.randint(0, 1000, size=(batch_size,)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, text_encoder_embedding_dim)).to(torch_device)
+        pooled_projections = torch.randn((batch_size, pooled_projection_dim)).to(torch_device)
+        encoder_attention_mask = torch.ones((batch_size, sequence_length)).to(torch_device)
+        guidance = torch.randint(0, 1000, size=(batch_size,)).to(torch_device, dtype=torch.float32)
+
+        return {
+            "hidden_states": hidden_states,
+            "timestep": timestep,
+            "encoder_hidden_states": encoder_hidden_states,
+            "pooled_projections": pooled_projections,
+            "encoder_attention_mask": encoder_attention_mask,
+            "guidance": guidance,
+        }
+
+    @property
+    def input_shape(self):
+        return (8, 1, 16, 16)
+
+    @property
+    def output_shape(self):
+        return (4, 1, 16, 16)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
+            "in_channels": 8,
+            "out_channels": 4,
+            "num_attention_heads": 2,
+            "attention_head_dim": 10,
+            "num_layers": 1,
+            "num_single_layers": 1,
+            "num_refiner_layers": 1,
+            "patch_size": 1,
+            "patch_size_t": 1,
+            "guidance_embeds": True,
+            "text_embed_dim": 16,
+            "pooled_projection_dim": 8,
+            "rope_axes_dim": (2, 4, 4),
+        }
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
+
+    def test_output(self):
+        super().test_output(expected_output_shape=(1, *self.output_shape))
+
+    def test_gradient_checkpointing_is_applied(self):
+        expected_set = {"HunyuanVideoTransformer3DModel"}
+        super().test_gradient_checkpointing_is_applied(expected_set=expected_set)

tests/pipelines/hunyuan_video/test_hunyuan_skyreels_image2video.py

+# Copyright 2024 The HuggingFace Team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import unittest
+
+import numpy as np
+import torch
+from PIL import Image
+from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer, LlamaConfig, LlamaModel, LlamaTokenizer
+
+from diffusers import (
+    AutoencoderKLHunyuanVideo,
+    FlowMatchEulerDiscreteScheduler,
+    HunyuanSkyreelsImageToVideoPipeline,
+    HunyuanVideoTransformer3DModel,
+)
+from diffusers.utils.testing_utils import enable_full_determinism, torch_device
+
+from ..test_pipelines_common import PipelineTesterMixin, PyramidAttentionBroadcastTesterMixin, to_np
+
+
+enable_full_determinism()
+
+
+class HunyuanSkyreelsImageToVideoPipelineFastTests(
+    PipelineTesterMixin, PyramidAttentionBroadcastTesterMixin, unittest.TestCase
+):
+    pipeline_class = HunyuanSkyreelsImageToVideoPipeline
+    params = frozenset(
+        ["image", "prompt", "height", "width", "guidance_scale", "prompt_embeds", "pooled_prompt_embeds"]
+    )
+    batch_params = frozenset(["prompt", "image"])
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    supports_dduf = False
+
+    # there is no xformers processor for Flux
+    test_xformers_attention = False
+    test_layerwise_casting = True
+    test_group_offloading = True
+
+    def get_dummy_components(self, num_layers: int = 1, num_single_layers: int = 1):
+        torch.manual_seed(0)
+        transformer = HunyuanVideoTransformer3DModel(
+            in_channels=8,
+            out_channels=4,
+            num_attention_heads=2,
+            attention_head_dim=10,
+            num_layers=num_layers,
+            num_single_layers=num_single_layers,
+            num_refiner_layers=1,
+            patch_size=1,
+            patch_size_t=1,
+            guidance_embeds=True,
+            text_embed_dim=16,
+            pooled_projection_dim=8,
+            rope_axes_dim=(2, 4, 4),
+        )
+
+        torch.manual_seed(0)
+        vae = AutoencoderKLHunyuanVideo(
+            in_channels=3,
+            out_channels=3,
+            latent_channels=4,
+            down_block_types=(
+                "HunyuanVideoDownBlock3D",
+                "HunyuanVideoDownBlock3D",
+                "HunyuanVideoDownBlock3D",
+                "HunyuanVideoDownBlock3D",
+            ),
+            up_block_types=(
+                "HunyuanVideoUpBlock3D",
+                "HunyuanVideoUpBlock3D",
+                "HunyuanVideoUpBlock3D",
+                "HunyuanVideoUpBlock3D",
+            ),
+            block_out_channels=(8, 8, 8, 8),
+            layers_per_block=1,
+            act_fn="silu",
+            norm_num_groups=4,
+            scaling_factor=0.476986,
+            spatial_compression_ratio=8,
+            temporal_compression_ratio=4,
+            mid_block_add_attention=True,
+        )
+
+        torch.manual_seed(0)
+        scheduler = FlowMatchEulerDiscreteScheduler(shift=7.0)
+
+        llama_text_encoder_config = LlamaConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=16,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=2,
+            pad_token_id=1,
+            vocab_size=1000,
+            hidden_act="gelu",
+            projection_dim=32,
+        )
+        clip_text_encoder_config = CLIPTextConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=8,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=2,
+            pad_token_id=1,
+            vocab_size=1000,
+            hidden_act="gelu",
+            projection_dim=32,
+        )
+
+        torch.manual_seed(0)
+        text_encoder = LlamaModel(llama_text_encoder_config)
+        tokenizer = LlamaTokenizer.from_pretrained("finetrainers/dummy-hunyaunvideo", subfolder="tokenizer")
+
+        torch.manual_seed(0)
+        text_encoder_2 = CLIPTextModel(clip_text_encoder_config)
+        tokenizer_2 = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "text_encoder_2": text_encoder_2,
+            "tokenizer": tokenizer,
+            "tokenizer_2": tokenizer_2,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+
+        image_height = 16
+        image_width = 16
+        image = Image.new("RGB", (image_width, image_height))
+        inputs = {
+            "image": image,
+            "prompt": "dance monkey",
+            "prompt_template": {
+                "template": "{}",
+                "crop_start": 0,
+            },
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 4.5,
+            "height": 16,
+            "width": 16,
+            # 4 * k + 1 is the recommendation
+            "num_frames": 9,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+
+        self.assertEqual(generated_video.shape, (9, 3, 16, 16))
+        expected_video = torch.randn(9, 3, 16, 16)
+        max_diff = np.abs(generated_video - expected_video).max()
+        self.assertLessEqual(max_diff, 1e10)
+
+    def test_callback_inputs(self):
+        sig = inspect.signature(self.pipeline_class.__call__)
+        has_callback_tensor_inputs = "callback_on_step_end_tensor_inputs" in sig.parameters
+        has_callback_step_end = "callback_on_step_end" in sig.parameters
+
+        if not (has_callback_tensor_inputs and has_callback_step_end):
+            return
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe = pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        self.assertTrue(
+            hasattr(pipe, "_callback_tensor_inputs"),
+            f" {self.pipeline_class} should have `_callback_tensor_inputs` that defines a list of tensor variables its callback function can use as inputs",
+        )
+
+        def callback_inputs_subset(pipe, i, t, callback_kwargs):
+            # iterate over callback args
+            for tensor_name, tensor_value in callback_kwargs.items():
+                # check that we're only passing in allowed tensor inputs
+                assert tensor_name in pipe._callback_tensor_inputs
+
+            return callback_kwargs
+
+        def callback_inputs_all(pipe, i, t, callback_kwargs):
+            for tensor_name in pipe._callback_tensor_inputs:
+                assert tensor_name in callback_kwargs
+
+            # iterate over callback args
+            for tensor_name, tensor_value in callback_kwargs.items():
+                # check that we're only passing in allowed tensor inputs
+                assert tensor_name in pipe._callback_tensor_inputs
+
+            return callback_kwargs
+
+        inputs = self.get_dummy_inputs(torch_device)
+
+        # Test passing in a subset
+        inputs["callback_on_step_end"] = callback_inputs_subset
+        inputs["callback_on_step_end_tensor_inputs"] = ["latents"]
+        output = pipe(**inputs)[0]
+
+        # Test passing in a everything
+        inputs["callback_on_step_end"] = callback_inputs_all
+        inputs["callback_on_step_end_tensor_inputs"] = pipe._callback_tensor_inputs
+        output = pipe(**inputs)[0]
+
+        def callback_inputs_change_tensor(pipe, i, t, callback_kwargs):
+            is_last = i == (pipe.num_timesteps - 1)
+            if is_last:
+                callback_kwargs["latents"] = torch.zeros_like(callback_kwargs["latents"])
+            return callback_kwargs
+
+        inputs["callback_on_step_end"] = callback_inputs_change_tensor
+        inputs["callback_on_step_end_tensor_inputs"] = pipe._callback_tensor_inputs
+        output = pipe(**inputs)[0]
+        assert output.abs().sum() < 1e10
+
+    def test_attention_slicing_forward_pass(
+        self, test_max_difference=True, test_mean_pixel_difference=True, expected_max_diff=1e-3
+    ):
+        if not self.test_attention_slicing:
+            return
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        output_without_slicing = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=1)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing1 = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=2)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing2 = pipe(**inputs)[0]
+
+        if test_max_difference:
+            max_diff1 = np.abs(to_np(output_with_slicing1) - to_np(output_without_slicing)).max()
+            max_diff2 = np.abs(to_np(output_with_slicing2) - to_np(output_without_slicing)).max()
+            self.assertLess(
+                max(max_diff1, max_diff2),
+                expected_max_diff,
+                "Attention slicing should not affect the inference results",
+            )
+
+    def test_vae_tiling(self, expected_diff_max: float = 0.2):
+        # Seems to require higher tolerance than the other tests
+        expected_diff_max = 0.6
+        generator_device = "cpu"
+        components = self.get_dummy_components()
+
+        pipe = self.pipeline_class(**components)
+        pipe.to("cpu")
+        pipe.set_progress_bar_config(disable=None)
+
+        # Without tiling
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_without_tiling = pipe(**inputs)[0]
+
+        # With tiling
+        pipe.vae.enable_tiling(
+            tile_sample_min_height=96,
+            tile_sample_min_width=96,
+            tile_sample_stride_height=64,
+            tile_sample_stride_width=64,
+        )
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_with_tiling = pipe(**inputs)[0]
+
+        self.assertLess(
+            (to_np(output_without_tiling) - to_np(output_with_tiling)).max(),
+            expected_diff_max,
+            "VAE tiling should not affect the inference results",
+        )
+
+    # TODO(aryan): Create a dummy gemma model with smol vocab size
+    @unittest.skip(
+        "A very small vocab size is used for fast tests. So, any kind of prompt other than the empty default used in other tests will lead to a embedding lookup error. This test uses a long prompt that causes the error."
+    )
+    def test_inference_batch_consistent(self):
+        pass
+
+    @unittest.skip(
+        "A very small vocab size is used for fast tests. So, any kind of prompt other than the empty default used in other tests will lead to a embedding lookup error. This test uses a long prompt that causes the error."
+    )
+    def test_inference_batch_single_identical(self):
+        pass